Physical inactivity increases the risk of many non-communicable diseases and has been recognised as a major contributor to the global burden of ill health [1]. It is therefore worrying that in many high-income settings average time spent in moderate to vigorous physical activity (MVPA) is well below the recommendation of 60 min per day [2]. A recent study in the UK, for example found that only 51% of 7 year olds met this target [3]. Understanding the barriers and drivers of physical activity in children is key to developing sustainable and successful interventions to increase activity levels. It has been consistently observed that children’s activity levels exhibit a seasonal pattern. This has been reported in many settings including Europe [4–12], the USA [13–15] and Australia [16]. Activity levels are generally lowest in the winter, when dark evenings and cool, wet weather is thought to inhibit activity [17]. Understanding how day length and weather influence physical activity is therefore a useful step in the development of sustainable interventions to maintain activity levels throughout the year.

Several studies have found relationships between different weather variables and children’s physical activity. Rainfall has been associated with decreased activity [18–22]. For example, Harrison and colleagues analysed a sample of 1794 9–10 year old English children and found they undertook almost 15 min less MVPA on the wettest days compared to days with no rain [20]. Conversely, temperature has shown a positive association with physical activity, with small to moderate increases in step counts associated with a 10 °C increase in temperature in studies in New Zealand (n = 1115) [21] and Canada (n = 1293) [18]. In addition to these relatively simple associations, there are suggestions that certain combinations of weather variables may have different associations with physical activity in adults [23], and that the relationships seen between weather conditions and physical activity may vary with age [20]. However, studies examining weather and physical activity have tended to be confined to small areas, often a single city or state, which limits variability in exposure measures. Although the weather varies daily within a single location, heterogeneity of weather exposures is often low in single site studies within a limited time frame. Further, the different means by which both physical activity and weather can be measured makes it difficult to compare their relationships with physical activity across different settings and populations. This is problematic because understanding if and how children in different settings respond to similar weather conditions might help us to identify potential adaptive strategies. For example, there may be settings where the physical environment supports outdoor active play in wet weather, or where the cultural environment promotes activity even in the cold. The fact that seasonal patterns in physical activity [24] and association with day length [25] are not consistently observed in all settings suggests that there may be some cultural adaptations to the markers of seasons; weather and day length.

The pooling of objectively measured physical activity from studies conducted around the world in the International Children’s Accelerometry Database (ICAD) provides the opportunity to explore the relationships between weather and physical activity over a wide range of exposures, and understand if and how these relationships differ by setting (e.g. the city/country in which studies are located) and participants.

This work was prompted in part by the observation that the seasonal patterns in children’s physical activity clearly visible in the UK are not necessarily replicated elsewhere [24]. Our analyses have shown that the relationships between weather conditions and physical activity vary between settings. It appears that children from Melbourne, Australia and Northern Europe that have higher activity levels given the weather conditions they experience compared to those in Western Europe and USA.

The use of a pooled dataset from around the world allowed us to get a clear picture of the relationship between physical activity and different weather variables. Temperature, precipitation, wind speed and visibility are all significantly associated with children’s physical activity across a wide range of exposures. While most weather conditions show broadly linear relationships with physical activity, temperature exhibits a more complex relationship, being associated with increased activity in the range 0–20 °C and decreased activity at higher temperatures. Past studies undertaken in temperate locations have typically reported a straightforward linear relationship [18, 21, 22] of increased temperature associated with increased physical activity. Our results suggest that where mean temperatures sit within the range of 0–20 °C that this is a reasonable assumption. However, we found that mean daily temperatures below 0 °C showed a flatter relationship with physical activity, and those higher than 20 °C were associated with a decline in physical activity. The latter association is similar to that seen in southeast Australia [34] and is likely to be a contributory factor to the reduced activity levels levels seen during high summer in another recent Australian study [35].

In addition to investigating the main effects of different weather conditions, we set out to establish if and how these factors interact with each other. Past research has associated overall climatic conditions with varying physical activity levels in adults [23]. However, we found relatively few significant interactions between the weather conditions, suggesting that combinations of conditions may generally be additive rather than interactive. Of the six interactions between the weather variables, only one was statistically significant (temperature and visibility), and there was one statistically significant interaction between weather and day length (visibility).

Age showed an interaction with two of the four weather variables. This included a significant interaction between age and precipitation. Among pre-school aged children the relationship appeared to be positive, with high rainfall associated with increased physical activity. Few studies have examined the relationships between weather conditions and physical activity among pre-school aged children. Where these associations have been studied, results are inconclusive [36], although there is some evidence that rainfall is associated with decreased physical activity [37]. This age group represents a small proportion of the ICAD sample, and is drawn from a small number of ICAD studies (eight of the 22 settings), so care must be taken in interpreting these findings, but it is suggestive that the impact of weather conditions in younger children’s physical activity may be different to that in older children and adolescents, and that these associations should be investigated further. It may also be instructive for future work to consider other non-climatic effect modifiers. Previous work in the UK has found an interaction between accelerometer-measured physical activity and week/weekend day whereby seasonal patterns are stronger on weekends [12]

Given the large size of our data set, it is perhaps unsurprising that most of the variables we included in our models show a statistically significant relationship with physical activity. But, importantly, the magnitude of the effect of these variables was comparable to that of established correlates of physical activity behaviour in these analyses. Using the fully adjusted model as given in Table 2 to predict cpm across the full ranges of the independent variables shows that a 2 cm increase in rainfall is associated with a slightly larger decrease in cpm (−26.4 cpm) than that associated with being overweight or obese relative to healthy weight (−24.4 cpm), or a year’s increase in age (−25.9 cpm).

Looking at the setting level random effects shows that some settings have higher average cpm after adjustment for individual level covariates and weather variables. In terms of regional trends, studies set in Northern Europe and Melbourne, Australia appear to better than average given the weather conditions they experience. Children in these regions are apparently more active on average (see Figs. 2 and 5), and when exposed to adverse weather conditions (Figs. 5 and 6). The six TAAG sites are all clustered towards the bottom of Fig. 6, indicating lower than average cpm given the characteristics of the participants and the weather conditions experienced. This could indicate a country level effect, the sites being spread across the United States, with potential cultural response to the weather. However, the Iowa Bone Development study, based in Des Moines, sits much further up the Figure.

It is perhaps particularly instructive to compare estimates for settings that appear to have similar weather conditions. Bristol, Norwich, Des Moines and Antwerp all had similar mean temperatures and rainfall, but are ranked quite differently in Fig. 6, with Bristol and Des Moines having higher than average (Bristol-ALSPAC and Des Moines), and average (Bristol-PEACH) cpm given the weather conditions they experience, and Norwich and Antwerp being below average. It could be that children in Bristol and Des Moines are better adapted in some way to these conditions, although differences in the effect between the two Bristol studies (ALSPAC and PEACH) suggest that these differences could be due to the differing samples rather than settings.

Our results have methodological implications for the collection of physical activity data and the comparison of such data between studies and places. Weather conditions are associated with children’s physical activity, and the associations differ between settings. Adjustment should therefore be made for both season and weather when comparing physical activity across settings, and over time. Although it has been suggested that weather and season need not be a concern in all settings in adult populations [38], our findings show that for children, especially pre-school and primary school-aged children, they should both be considered in all settings. The nature of their relationships with children’s physical activity may vary with setting and exposure range, but their impact should be locally investigated before they are ruled out as determinants of behaviour.

This study has a number of strengths and limitations. We were able to use a large, international dataset for which physical activity was measured objectively. The pooling of data within ICAD provided consistent accelerometer processing protocols and weather data were extracted from a similarly standardised data source. The international scope of both datasets allowed the examination of weather and physical activity relationships across a range of locations and therefore a large range of exposures.

In terms of limitations, weather data for each day in each setting were derived from a single weather station and the link between physical activity and weather had to be made on the basis of overall study location rather than participant home location. Although this is a method used in many studies investigating relationships between weather and behaviour [18–22, 25, 39], it limits the precision of weather exposure estimates, and may lead to some attenuation of associations with physical activity. The ICAD data were recorded in a wide range of weather conditions. However the different study samples, and the different times of year over which they were collected, meant that these exposures were not equally distributed across different ages and sexes, and that in some locations measurements were taken over smaller portions of the year than others. For example, some of the warmer study locations were centres in the TAAG study, which investigated physical activity in adolescent girls only. Although our multilevel modelling approach adjusts for these differences to some extent, some of the setting level differences we observe may reflect differences between samples rather than locations. Some studies were based in schools, and we were not able to account for potential school and class level clustering, for example.

Although accelerometers provide an objective measure of physical activity they are not without their limitations. They have a limited ability to assess activity while the wearer is cycling [40], and the units used in ICAD must be removed altogether during aquatic activities. This may have impacted the associations we observed if children’s propensity to undertake these types of activity is associated with weather conditions. Finally, the studies included in these analyses all come from high-income countries, meaning our findings may not be applicable in other countries.

We find variation in the relationship between weather conditions and physical activity between ICAD studies and settings. There is evidence that children in Northern Europe and Melbourne, Australia are not only more active on average, but also more active given the weather conditions they experience. Further investigations into the nature of physical activity in these regions may therefore be useful to develop strategies to promote sustainable physical activity increases elsewhere. Weather conditions undoubtedly appear to present a barrier to physical activity, particularly among young children. Future work should therefore consider strategies to mitigate such impacts, and interventions involving changes to the physical environment should consider how they will operate in different weather conditions.